xref: /openbmc/linux/fs/super.c (revision 174cd4b1)
1  /*
2   *  linux/fs/super.c
3   *
4   *  Copyright (C) 1991, 1992  Linus Torvalds
5   *
6   *  super.c contains code to handle: - mount structures
7   *                                   - super-block tables
8   *                                   - filesystem drivers list
9   *                                   - mount system call
10   *                                   - umount system call
11   *                                   - ustat system call
12   *
13   * GK 2/5/95  -  Changed to support mounting the root fs via NFS
14   *
15   *  Added kerneld support: Jacques Gelinas and Bjorn Ekwall
16   *  Added change_root: Werner Almesberger & Hans Lermen, Feb '96
17   *  Added options to /proc/mounts:
18   *    Torbjörn Lindh (torbjorn.lindh@gopta.se), April 14, 1996.
19   *  Added devfs support: Richard Gooch <rgooch@atnf.csiro.au>, 13-JAN-1998
20   *  Heavily rewritten for 'one fs - one tree' dcache architecture. AV, Mar 2000
21   */
22  
23  #include <linux/export.h>
24  #include <linux/slab.h>
25  #include <linux/blkdev.h>
26  #include <linux/mount.h>
27  #include <linux/security.h>
28  #include <linux/writeback.h>		/* for the emergency remount stuff */
29  #include <linux/idr.h>
30  #include <linux/mutex.h>
31  #include <linux/backing-dev.h>
32  #include <linux/rculist_bl.h>
33  #include <linux/cleancache.h>
34  #include <linux/fsnotify.h>
35  #include <linux/lockdep.h>
36  #include <linux/user_namespace.h>
37  #include "internal.h"
38  
39  
40  static LIST_HEAD(super_blocks);
41  static DEFINE_SPINLOCK(sb_lock);
42  
43  static char *sb_writers_name[SB_FREEZE_LEVELS] = {
44  	"sb_writers",
45  	"sb_pagefaults",
46  	"sb_internal",
47  };
48  
49  /*
50   * One thing we have to be careful of with a per-sb shrinker is that we don't
51   * drop the last active reference to the superblock from within the shrinker.
52   * If that happens we could trigger unregistering the shrinker from within the
53   * shrinker path and that leads to deadlock on the shrinker_rwsem. Hence we
54   * take a passive reference to the superblock to avoid this from occurring.
55   */
56  static unsigned long super_cache_scan(struct shrinker *shrink,
57  				      struct shrink_control *sc)
58  {
59  	struct super_block *sb;
60  	long	fs_objects = 0;
61  	long	total_objects;
62  	long	freed = 0;
63  	long	dentries;
64  	long	inodes;
65  
66  	sb = container_of(shrink, struct super_block, s_shrink);
67  
68  	/*
69  	 * Deadlock avoidance.  We may hold various FS locks, and we don't want
70  	 * to recurse into the FS that called us in clear_inode() and friends..
71  	 */
72  	if (!(sc->gfp_mask & __GFP_FS))
73  		return SHRINK_STOP;
74  
75  	if (!trylock_super(sb))
76  		return SHRINK_STOP;
77  
78  	if (sb->s_op->nr_cached_objects)
79  		fs_objects = sb->s_op->nr_cached_objects(sb, sc);
80  
81  	inodes = list_lru_shrink_count(&sb->s_inode_lru, sc);
82  	dentries = list_lru_shrink_count(&sb->s_dentry_lru, sc);
83  	total_objects = dentries + inodes + fs_objects + 1;
84  	if (!total_objects)
85  		total_objects = 1;
86  
87  	/* proportion the scan between the caches */
88  	dentries = mult_frac(sc->nr_to_scan, dentries, total_objects);
89  	inodes = mult_frac(sc->nr_to_scan, inodes, total_objects);
90  	fs_objects = mult_frac(sc->nr_to_scan, fs_objects, total_objects);
91  
92  	/*
93  	 * prune the dcache first as the icache is pinned by it, then
94  	 * prune the icache, followed by the filesystem specific caches
95  	 *
96  	 * Ensure that we always scan at least one object - memcg kmem
97  	 * accounting uses this to fully empty the caches.
98  	 */
99  	sc->nr_to_scan = dentries + 1;
100  	freed = prune_dcache_sb(sb, sc);
101  	sc->nr_to_scan = inodes + 1;
102  	freed += prune_icache_sb(sb, sc);
103  
104  	if (fs_objects) {
105  		sc->nr_to_scan = fs_objects + 1;
106  		freed += sb->s_op->free_cached_objects(sb, sc);
107  	}
108  
109  	up_read(&sb->s_umount);
110  	return freed;
111  }
112  
113  static unsigned long super_cache_count(struct shrinker *shrink,
114  				       struct shrink_control *sc)
115  {
116  	struct super_block *sb;
117  	long	total_objects = 0;
118  
119  	sb = container_of(shrink, struct super_block, s_shrink);
120  
121  	/*
122  	 * Don't call trylock_super as it is a potential
123  	 * scalability bottleneck. The counts could get updated
124  	 * between super_cache_count and super_cache_scan anyway.
125  	 * Call to super_cache_count with shrinker_rwsem held
126  	 * ensures the safety of call to list_lru_shrink_count() and
127  	 * s_op->nr_cached_objects().
128  	 */
129  	if (sb->s_op && sb->s_op->nr_cached_objects)
130  		total_objects = sb->s_op->nr_cached_objects(sb, sc);
131  
132  	total_objects += list_lru_shrink_count(&sb->s_dentry_lru, sc);
133  	total_objects += list_lru_shrink_count(&sb->s_inode_lru, sc);
134  
135  	total_objects = vfs_pressure_ratio(total_objects);
136  	return total_objects;
137  }
138  
139  static void destroy_super_work(struct work_struct *work)
140  {
141  	struct super_block *s = container_of(work, struct super_block,
142  							destroy_work);
143  	int i;
144  
145  	for (i = 0; i < SB_FREEZE_LEVELS; i++)
146  		percpu_free_rwsem(&s->s_writers.rw_sem[i]);
147  	kfree(s);
148  }
149  
150  static void destroy_super_rcu(struct rcu_head *head)
151  {
152  	struct super_block *s = container_of(head, struct super_block, rcu);
153  	INIT_WORK(&s->destroy_work, destroy_super_work);
154  	schedule_work(&s->destroy_work);
155  }
156  
157  /**
158   *	destroy_super	-	frees a superblock
159   *	@s: superblock to free
160   *
161   *	Frees a superblock.
162   */
163  static void destroy_super(struct super_block *s)
164  {
165  	list_lru_destroy(&s->s_dentry_lru);
166  	list_lru_destroy(&s->s_inode_lru);
167  	security_sb_free(s);
168  	WARN_ON(!list_empty(&s->s_mounts));
169  	put_user_ns(s->s_user_ns);
170  	kfree(s->s_subtype);
171  	kfree(s->s_options);
172  	call_rcu(&s->rcu, destroy_super_rcu);
173  }
174  
175  /**
176   *	alloc_super	-	create new superblock
177   *	@type:	filesystem type superblock should belong to
178   *	@flags: the mount flags
179   *	@user_ns: User namespace for the super_block
180   *
181   *	Allocates and initializes a new &struct super_block.  alloc_super()
182   *	returns a pointer new superblock or %NULL if allocation had failed.
183   */
184  static struct super_block *alloc_super(struct file_system_type *type, int flags,
185  				       struct user_namespace *user_ns)
186  {
187  	struct super_block *s = kzalloc(sizeof(struct super_block),  GFP_USER);
188  	static const struct super_operations default_op;
189  	int i;
190  
191  	if (!s)
192  		return NULL;
193  
194  	INIT_LIST_HEAD(&s->s_mounts);
195  	s->s_user_ns = get_user_ns(user_ns);
196  
197  	if (security_sb_alloc(s))
198  		goto fail;
199  
200  	for (i = 0; i < SB_FREEZE_LEVELS; i++) {
201  		if (__percpu_init_rwsem(&s->s_writers.rw_sem[i],
202  					sb_writers_name[i],
203  					&type->s_writers_key[i]))
204  			goto fail;
205  	}
206  	init_waitqueue_head(&s->s_writers.wait_unfrozen);
207  	s->s_bdi = &noop_backing_dev_info;
208  	s->s_flags = flags;
209  	if (s->s_user_ns != &init_user_ns)
210  		s->s_iflags |= SB_I_NODEV;
211  	INIT_HLIST_NODE(&s->s_instances);
212  	INIT_HLIST_BL_HEAD(&s->s_anon);
213  	mutex_init(&s->s_sync_lock);
214  	INIT_LIST_HEAD(&s->s_inodes);
215  	spin_lock_init(&s->s_inode_list_lock);
216  	INIT_LIST_HEAD(&s->s_inodes_wb);
217  	spin_lock_init(&s->s_inode_wblist_lock);
218  
219  	if (list_lru_init_memcg(&s->s_dentry_lru))
220  		goto fail;
221  	if (list_lru_init_memcg(&s->s_inode_lru))
222  		goto fail;
223  
224  	init_rwsem(&s->s_umount);
225  	lockdep_set_class(&s->s_umount, &type->s_umount_key);
226  	/*
227  	 * sget() can have s_umount recursion.
228  	 *
229  	 * When it cannot find a suitable sb, it allocates a new
230  	 * one (this one), and tries again to find a suitable old
231  	 * one.
232  	 *
233  	 * In case that succeeds, it will acquire the s_umount
234  	 * lock of the old one. Since these are clearly distrinct
235  	 * locks, and this object isn't exposed yet, there's no
236  	 * risk of deadlocks.
237  	 *
238  	 * Annotate this by putting this lock in a different
239  	 * subclass.
240  	 */
241  	down_write_nested(&s->s_umount, SINGLE_DEPTH_NESTING);
242  	s->s_count = 1;
243  	atomic_set(&s->s_active, 1);
244  	mutex_init(&s->s_vfs_rename_mutex);
245  	lockdep_set_class(&s->s_vfs_rename_mutex, &type->s_vfs_rename_key);
246  	mutex_init(&s->s_dquot.dqio_mutex);
247  	s->s_maxbytes = MAX_NON_LFS;
248  	s->s_op = &default_op;
249  	s->s_time_gran = 1000000000;
250  	s->cleancache_poolid = CLEANCACHE_NO_POOL;
251  
252  	s->s_shrink.seeks = DEFAULT_SEEKS;
253  	s->s_shrink.scan_objects = super_cache_scan;
254  	s->s_shrink.count_objects = super_cache_count;
255  	s->s_shrink.batch = 1024;
256  	s->s_shrink.flags = SHRINKER_NUMA_AWARE | SHRINKER_MEMCG_AWARE;
257  	return s;
258  
259  fail:
260  	destroy_super(s);
261  	return NULL;
262  }
263  
264  /* Superblock refcounting  */
265  
266  /*
267   * Drop a superblock's refcount.  The caller must hold sb_lock.
268   */
269  static void __put_super(struct super_block *sb)
270  {
271  	if (!--sb->s_count) {
272  		list_del_init(&sb->s_list);
273  		destroy_super(sb);
274  	}
275  }
276  
277  /**
278   *	put_super	-	drop a temporary reference to superblock
279   *	@sb: superblock in question
280   *
281   *	Drops a temporary reference, frees superblock if there's no
282   *	references left.
283   */
284  static void put_super(struct super_block *sb)
285  {
286  	spin_lock(&sb_lock);
287  	__put_super(sb);
288  	spin_unlock(&sb_lock);
289  }
290  
291  
292  /**
293   *	deactivate_locked_super	-	drop an active reference to superblock
294   *	@s: superblock to deactivate
295   *
296   *	Drops an active reference to superblock, converting it into a temporary
297   *	one if there is no other active references left.  In that case we
298   *	tell fs driver to shut it down and drop the temporary reference we
299   *	had just acquired.
300   *
301   *	Caller holds exclusive lock on superblock; that lock is released.
302   */
303  void deactivate_locked_super(struct super_block *s)
304  {
305  	struct file_system_type *fs = s->s_type;
306  	if (atomic_dec_and_test(&s->s_active)) {
307  		cleancache_invalidate_fs(s);
308  		unregister_shrinker(&s->s_shrink);
309  		fs->kill_sb(s);
310  
311  		/*
312  		 * Since list_lru_destroy() may sleep, we cannot call it from
313  		 * put_super(), where we hold the sb_lock. Therefore we destroy
314  		 * the lru lists right now.
315  		 */
316  		list_lru_destroy(&s->s_dentry_lru);
317  		list_lru_destroy(&s->s_inode_lru);
318  
319  		put_filesystem(fs);
320  		put_super(s);
321  	} else {
322  		up_write(&s->s_umount);
323  	}
324  }
325  
326  EXPORT_SYMBOL(deactivate_locked_super);
327  
328  /**
329   *	deactivate_super	-	drop an active reference to superblock
330   *	@s: superblock to deactivate
331   *
332   *	Variant of deactivate_locked_super(), except that superblock is *not*
333   *	locked by caller.  If we are going to drop the final active reference,
334   *	lock will be acquired prior to that.
335   */
336  void deactivate_super(struct super_block *s)
337  {
338          if (!atomic_add_unless(&s->s_active, -1, 1)) {
339  		down_write(&s->s_umount);
340  		deactivate_locked_super(s);
341  	}
342  }
343  
344  EXPORT_SYMBOL(deactivate_super);
345  
346  /**
347   *	grab_super - acquire an active reference
348   *	@s: reference we are trying to make active
349   *
350   *	Tries to acquire an active reference.  grab_super() is used when we
351   * 	had just found a superblock in super_blocks or fs_type->fs_supers
352   *	and want to turn it into a full-blown active reference.  grab_super()
353   *	is called with sb_lock held and drops it.  Returns 1 in case of
354   *	success, 0 if we had failed (superblock contents was already dead or
355   *	dying when grab_super() had been called).  Note that this is only
356   *	called for superblocks not in rundown mode (== ones still on ->fs_supers
357   *	of their type), so increment of ->s_count is OK here.
358   */
359  static int grab_super(struct super_block *s) __releases(sb_lock)
360  {
361  	s->s_count++;
362  	spin_unlock(&sb_lock);
363  	down_write(&s->s_umount);
364  	if ((s->s_flags & MS_BORN) && atomic_inc_not_zero(&s->s_active)) {
365  		put_super(s);
366  		return 1;
367  	}
368  	up_write(&s->s_umount);
369  	put_super(s);
370  	return 0;
371  }
372  
373  /*
374   *	trylock_super - try to grab ->s_umount shared
375   *	@sb: reference we are trying to grab
376   *
377   *	Try to prevent fs shutdown.  This is used in places where we
378   *	cannot take an active reference but we need to ensure that the
379   *	filesystem is not shut down while we are working on it. It returns
380   *	false if we cannot acquire s_umount or if we lose the race and
381   *	filesystem already got into shutdown, and returns true with the s_umount
382   *	lock held in read mode in case of success. On successful return,
383   *	the caller must drop the s_umount lock when done.
384   *
385   *	Note that unlike get_super() et.al. this one does *not* bump ->s_count.
386   *	The reason why it's safe is that we are OK with doing trylock instead
387   *	of down_read().  There's a couple of places that are OK with that, but
388   *	it's very much not a general-purpose interface.
389   */
390  bool trylock_super(struct super_block *sb)
391  {
392  	if (down_read_trylock(&sb->s_umount)) {
393  		if (!hlist_unhashed(&sb->s_instances) &&
394  		    sb->s_root && (sb->s_flags & MS_BORN))
395  			return true;
396  		up_read(&sb->s_umount);
397  	}
398  
399  	return false;
400  }
401  
402  /**
403   *	generic_shutdown_super	-	common helper for ->kill_sb()
404   *	@sb: superblock to kill
405   *
406   *	generic_shutdown_super() does all fs-independent work on superblock
407   *	shutdown.  Typical ->kill_sb() should pick all fs-specific objects
408   *	that need destruction out of superblock, call generic_shutdown_super()
409   *	and release aforementioned objects.  Note: dentries and inodes _are_
410   *	taken care of and do not need specific handling.
411   *
412   *	Upon calling this function, the filesystem may no longer alter or
413   *	rearrange the set of dentries belonging to this super_block, nor may it
414   *	change the attachments of dentries to inodes.
415   */
416  void generic_shutdown_super(struct super_block *sb)
417  {
418  	const struct super_operations *sop = sb->s_op;
419  
420  	if (sb->s_root) {
421  		shrink_dcache_for_umount(sb);
422  		sync_filesystem(sb);
423  		sb->s_flags &= ~MS_ACTIVE;
424  
425  		fsnotify_unmount_inodes(sb);
426  		cgroup_writeback_umount();
427  
428  		evict_inodes(sb);
429  
430  		if (sb->s_dio_done_wq) {
431  			destroy_workqueue(sb->s_dio_done_wq);
432  			sb->s_dio_done_wq = NULL;
433  		}
434  
435  		if (sop->put_super)
436  			sop->put_super(sb);
437  
438  		if (!list_empty(&sb->s_inodes)) {
439  			printk("VFS: Busy inodes after unmount of %s. "
440  			   "Self-destruct in 5 seconds.  Have a nice day...\n",
441  			   sb->s_id);
442  		}
443  	}
444  	spin_lock(&sb_lock);
445  	/* should be initialized for __put_super_and_need_restart() */
446  	hlist_del_init(&sb->s_instances);
447  	spin_unlock(&sb_lock);
448  	up_write(&sb->s_umount);
449  }
450  
451  EXPORT_SYMBOL(generic_shutdown_super);
452  
453  /**
454   *	sget_userns -	find or create a superblock
455   *	@type:	filesystem type superblock should belong to
456   *	@test:	comparison callback
457   *	@set:	setup callback
458   *	@flags:	mount flags
459   *	@user_ns: User namespace for the super_block
460   *	@data:	argument to each of them
461   */
462  struct super_block *sget_userns(struct file_system_type *type,
463  			int (*test)(struct super_block *,void *),
464  			int (*set)(struct super_block *,void *),
465  			int flags, struct user_namespace *user_ns,
466  			void *data)
467  {
468  	struct super_block *s = NULL;
469  	struct super_block *old;
470  	int err;
471  
472  	if (!(flags & (MS_KERNMOUNT|MS_SUBMOUNT)) &&
473  	    !(type->fs_flags & FS_USERNS_MOUNT) &&
474  	    !capable(CAP_SYS_ADMIN))
475  		return ERR_PTR(-EPERM);
476  retry:
477  	spin_lock(&sb_lock);
478  	if (test) {
479  		hlist_for_each_entry(old, &type->fs_supers, s_instances) {
480  			if (!test(old, data))
481  				continue;
482  			if (user_ns != old->s_user_ns) {
483  				spin_unlock(&sb_lock);
484  				if (s) {
485  					up_write(&s->s_umount);
486  					destroy_super(s);
487  				}
488  				return ERR_PTR(-EBUSY);
489  			}
490  			if (!grab_super(old))
491  				goto retry;
492  			if (s) {
493  				up_write(&s->s_umount);
494  				destroy_super(s);
495  				s = NULL;
496  			}
497  			return old;
498  		}
499  	}
500  	if (!s) {
501  		spin_unlock(&sb_lock);
502  		s = alloc_super(type, (flags & ~MS_SUBMOUNT), user_ns);
503  		if (!s)
504  			return ERR_PTR(-ENOMEM);
505  		goto retry;
506  	}
507  
508  	err = set(s, data);
509  	if (err) {
510  		spin_unlock(&sb_lock);
511  		up_write(&s->s_umount);
512  		destroy_super(s);
513  		return ERR_PTR(err);
514  	}
515  	s->s_type = type;
516  	strlcpy(s->s_id, type->name, sizeof(s->s_id));
517  	list_add_tail(&s->s_list, &super_blocks);
518  	hlist_add_head(&s->s_instances, &type->fs_supers);
519  	spin_unlock(&sb_lock);
520  	get_filesystem(type);
521  	register_shrinker(&s->s_shrink);
522  	return s;
523  }
524  
525  EXPORT_SYMBOL(sget_userns);
526  
527  /**
528   *	sget	-	find or create a superblock
529   *	@type:	  filesystem type superblock should belong to
530   *	@test:	  comparison callback
531   *	@set:	  setup callback
532   *	@flags:	  mount flags
533   *	@data:	  argument to each of them
534   */
535  struct super_block *sget(struct file_system_type *type,
536  			int (*test)(struct super_block *,void *),
537  			int (*set)(struct super_block *,void *),
538  			int flags,
539  			void *data)
540  {
541  	struct user_namespace *user_ns = current_user_ns();
542  
543  	/* We don't yet pass the user namespace of the parent
544  	 * mount through to here so always use &init_user_ns
545  	 * until that changes.
546  	 */
547  	if (flags & MS_SUBMOUNT)
548  		user_ns = &init_user_ns;
549  
550  	/* Ensure the requestor has permissions over the target filesystem */
551  	if (!(flags & (MS_KERNMOUNT|MS_SUBMOUNT)) && !ns_capable(user_ns, CAP_SYS_ADMIN))
552  		return ERR_PTR(-EPERM);
553  
554  	return sget_userns(type, test, set, flags, user_ns, data);
555  }
556  
557  EXPORT_SYMBOL(sget);
558  
559  void drop_super(struct super_block *sb)
560  {
561  	up_read(&sb->s_umount);
562  	put_super(sb);
563  }
564  
565  EXPORT_SYMBOL(drop_super);
566  
567  void drop_super_exclusive(struct super_block *sb)
568  {
569  	up_write(&sb->s_umount);
570  	put_super(sb);
571  }
572  EXPORT_SYMBOL(drop_super_exclusive);
573  
574  /**
575   *	iterate_supers - call function for all active superblocks
576   *	@f: function to call
577   *	@arg: argument to pass to it
578   *
579   *	Scans the superblock list and calls given function, passing it
580   *	locked superblock and given argument.
581   */
582  void iterate_supers(void (*f)(struct super_block *, void *), void *arg)
583  {
584  	struct super_block *sb, *p = NULL;
585  
586  	spin_lock(&sb_lock);
587  	list_for_each_entry(sb, &super_blocks, s_list) {
588  		if (hlist_unhashed(&sb->s_instances))
589  			continue;
590  		sb->s_count++;
591  		spin_unlock(&sb_lock);
592  
593  		down_read(&sb->s_umount);
594  		if (sb->s_root && (sb->s_flags & MS_BORN))
595  			f(sb, arg);
596  		up_read(&sb->s_umount);
597  
598  		spin_lock(&sb_lock);
599  		if (p)
600  			__put_super(p);
601  		p = sb;
602  	}
603  	if (p)
604  		__put_super(p);
605  	spin_unlock(&sb_lock);
606  }
607  
608  /**
609   *	iterate_supers_type - call function for superblocks of given type
610   *	@type: fs type
611   *	@f: function to call
612   *	@arg: argument to pass to it
613   *
614   *	Scans the superblock list and calls given function, passing it
615   *	locked superblock and given argument.
616   */
617  void iterate_supers_type(struct file_system_type *type,
618  	void (*f)(struct super_block *, void *), void *arg)
619  {
620  	struct super_block *sb, *p = NULL;
621  
622  	spin_lock(&sb_lock);
623  	hlist_for_each_entry(sb, &type->fs_supers, s_instances) {
624  		sb->s_count++;
625  		spin_unlock(&sb_lock);
626  
627  		down_read(&sb->s_umount);
628  		if (sb->s_root && (sb->s_flags & MS_BORN))
629  			f(sb, arg);
630  		up_read(&sb->s_umount);
631  
632  		spin_lock(&sb_lock);
633  		if (p)
634  			__put_super(p);
635  		p = sb;
636  	}
637  	if (p)
638  		__put_super(p);
639  	spin_unlock(&sb_lock);
640  }
641  
642  EXPORT_SYMBOL(iterate_supers_type);
643  
644  static struct super_block *__get_super(struct block_device *bdev, bool excl)
645  {
646  	struct super_block *sb;
647  
648  	if (!bdev)
649  		return NULL;
650  
651  	spin_lock(&sb_lock);
652  rescan:
653  	list_for_each_entry(sb, &super_blocks, s_list) {
654  		if (hlist_unhashed(&sb->s_instances))
655  			continue;
656  		if (sb->s_bdev == bdev) {
657  			sb->s_count++;
658  			spin_unlock(&sb_lock);
659  			if (!excl)
660  				down_read(&sb->s_umount);
661  			else
662  				down_write(&sb->s_umount);
663  			/* still alive? */
664  			if (sb->s_root && (sb->s_flags & MS_BORN))
665  				return sb;
666  			if (!excl)
667  				up_read(&sb->s_umount);
668  			else
669  				up_write(&sb->s_umount);
670  			/* nope, got unmounted */
671  			spin_lock(&sb_lock);
672  			__put_super(sb);
673  			goto rescan;
674  		}
675  	}
676  	spin_unlock(&sb_lock);
677  	return NULL;
678  }
679  
680  /**
681   *	get_super - get the superblock of a device
682   *	@bdev: device to get the superblock for
683   *
684   *	Scans the superblock list and finds the superblock of the file system
685   *	mounted on the device given. %NULL is returned if no match is found.
686   */
687  struct super_block *get_super(struct block_device *bdev)
688  {
689  	return __get_super(bdev, false);
690  }
691  EXPORT_SYMBOL(get_super);
692  
693  static struct super_block *__get_super_thawed(struct block_device *bdev,
694  					      bool excl)
695  {
696  	while (1) {
697  		struct super_block *s = __get_super(bdev, excl);
698  		if (!s || s->s_writers.frozen == SB_UNFROZEN)
699  			return s;
700  		if (!excl)
701  			up_read(&s->s_umount);
702  		else
703  			up_write(&s->s_umount);
704  		wait_event(s->s_writers.wait_unfrozen,
705  			   s->s_writers.frozen == SB_UNFROZEN);
706  		put_super(s);
707  	}
708  }
709  
710  /**
711   *	get_super_thawed - get thawed superblock of a device
712   *	@bdev: device to get the superblock for
713   *
714   *	Scans the superblock list and finds the superblock of the file system
715   *	mounted on the device. The superblock is returned once it is thawed
716   *	(or immediately if it was not frozen). %NULL is returned if no match
717   *	is found.
718   */
719  struct super_block *get_super_thawed(struct block_device *bdev)
720  {
721  	return __get_super_thawed(bdev, false);
722  }
723  EXPORT_SYMBOL(get_super_thawed);
724  
725  /**
726   *	get_super_exclusive_thawed - get thawed superblock of a device
727   *	@bdev: device to get the superblock for
728   *
729   *	Scans the superblock list and finds the superblock of the file system
730   *	mounted on the device. The superblock is returned once it is thawed
731   *	(or immediately if it was not frozen) and s_umount semaphore is held
732   *	in exclusive mode. %NULL is returned if no match is found.
733   */
734  struct super_block *get_super_exclusive_thawed(struct block_device *bdev)
735  {
736  	return __get_super_thawed(bdev, true);
737  }
738  EXPORT_SYMBOL(get_super_exclusive_thawed);
739  
740  /**
741   * get_active_super - get an active reference to the superblock of a device
742   * @bdev: device to get the superblock for
743   *
744   * Scans the superblock list and finds the superblock of the file system
745   * mounted on the device given.  Returns the superblock with an active
746   * reference or %NULL if none was found.
747   */
748  struct super_block *get_active_super(struct block_device *bdev)
749  {
750  	struct super_block *sb;
751  
752  	if (!bdev)
753  		return NULL;
754  
755  restart:
756  	spin_lock(&sb_lock);
757  	list_for_each_entry(sb, &super_blocks, s_list) {
758  		if (hlist_unhashed(&sb->s_instances))
759  			continue;
760  		if (sb->s_bdev == bdev) {
761  			if (!grab_super(sb))
762  				goto restart;
763  			up_write(&sb->s_umount);
764  			return sb;
765  		}
766  	}
767  	spin_unlock(&sb_lock);
768  	return NULL;
769  }
770  
771  struct super_block *user_get_super(dev_t dev)
772  {
773  	struct super_block *sb;
774  
775  	spin_lock(&sb_lock);
776  rescan:
777  	list_for_each_entry(sb, &super_blocks, s_list) {
778  		if (hlist_unhashed(&sb->s_instances))
779  			continue;
780  		if (sb->s_dev ==  dev) {
781  			sb->s_count++;
782  			spin_unlock(&sb_lock);
783  			down_read(&sb->s_umount);
784  			/* still alive? */
785  			if (sb->s_root && (sb->s_flags & MS_BORN))
786  				return sb;
787  			up_read(&sb->s_umount);
788  			/* nope, got unmounted */
789  			spin_lock(&sb_lock);
790  			__put_super(sb);
791  			goto rescan;
792  		}
793  	}
794  	spin_unlock(&sb_lock);
795  	return NULL;
796  }
797  
798  /**
799   *	do_remount_sb - asks filesystem to change mount options.
800   *	@sb:	superblock in question
801   *	@flags:	numeric part of options
802   *	@data:	the rest of options
803   *      @force: whether or not to force the change
804   *
805   *	Alters the mount options of a mounted file system.
806   */
807  int do_remount_sb(struct super_block *sb, int flags, void *data, int force)
808  {
809  	int retval;
810  	int remount_ro;
811  
812  	if (sb->s_writers.frozen != SB_UNFROZEN)
813  		return -EBUSY;
814  
815  #ifdef CONFIG_BLOCK
816  	if (!(flags & MS_RDONLY) && bdev_read_only(sb->s_bdev))
817  		return -EACCES;
818  #endif
819  
820  	remount_ro = (flags & MS_RDONLY) && !(sb->s_flags & MS_RDONLY);
821  
822  	if (remount_ro) {
823  		if (!hlist_empty(&sb->s_pins)) {
824  			up_write(&sb->s_umount);
825  			group_pin_kill(&sb->s_pins);
826  			down_write(&sb->s_umount);
827  			if (!sb->s_root)
828  				return 0;
829  			if (sb->s_writers.frozen != SB_UNFROZEN)
830  				return -EBUSY;
831  			remount_ro = (flags & MS_RDONLY) && !(sb->s_flags & MS_RDONLY);
832  		}
833  	}
834  	shrink_dcache_sb(sb);
835  
836  	/* If we are remounting RDONLY and current sb is read/write,
837  	   make sure there are no rw files opened */
838  	if (remount_ro) {
839  		if (force) {
840  			sb->s_readonly_remount = 1;
841  			smp_wmb();
842  		} else {
843  			retval = sb_prepare_remount_readonly(sb);
844  			if (retval)
845  				return retval;
846  		}
847  	}
848  
849  	if (sb->s_op->remount_fs) {
850  		retval = sb->s_op->remount_fs(sb, &flags, data);
851  		if (retval) {
852  			if (!force)
853  				goto cancel_readonly;
854  			/* If forced remount, go ahead despite any errors */
855  			WARN(1, "forced remount of a %s fs returned %i\n",
856  			     sb->s_type->name, retval);
857  		}
858  	}
859  	sb->s_flags = (sb->s_flags & ~MS_RMT_MASK) | (flags & MS_RMT_MASK);
860  	/* Needs to be ordered wrt mnt_is_readonly() */
861  	smp_wmb();
862  	sb->s_readonly_remount = 0;
863  
864  	/*
865  	 * Some filesystems modify their metadata via some other path than the
866  	 * bdev buffer cache (eg. use a private mapping, or directories in
867  	 * pagecache, etc). Also file data modifications go via their own
868  	 * mappings. So If we try to mount readonly then copy the filesystem
869  	 * from bdev, we could get stale data, so invalidate it to give a best
870  	 * effort at coherency.
871  	 */
872  	if (remount_ro && sb->s_bdev)
873  		invalidate_bdev(sb->s_bdev);
874  	return 0;
875  
876  cancel_readonly:
877  	sb->s_readonly_remount = 0;
878  	return retval;
879  }
880  
881  static void do_emergency_remount(struct work_struct *work)
882  {
883  	struct super_block *sb, *p = NULL;
884  
885  	spin_lock(&sb_lock);
886  	list_for_each_entry(sb, &super_blocks, s_list) {
887  		if (hlist_unhashed(&sb->s_instances))
888  			continue;
889  		sb->s_count++;
890  		spin_unlock(&sb_lock);
891  		down_write(&sb->s_umount);
892  		if (sb->s_root && sb->s_bdev && (sb->s_flags & MS_BORN) &&
893  		    !(sb->s_flags & MS_RDONLY)) {
894  			/*
895  			 * What lock protects sb->s_flags??
896  			 */
897  			do_remount_sb(sb, MS_RDONLY, NULL, 1);
898  		}
899  		up_write(&sb->s_umount);
900  		spin_lock(&sb_lock);
901  		if (p)
902  			__put_super(p);
903  		p = sb;
904  	}
905  	if (p)
906  		__put_super(p);
907  	spin_unlock(&sb_lock);
908  	kfree(work);
909  	printk("Emergency Remount complete\n");
910  }
911  
912  void emergency_remount(void)
913  {
914  	struct work_struct *work;
915  
916  	work = kmalloc(sizeof(*work), GFP_ATOMIC);
917  	if (work) {
918  		INIT_WORK(work, do_emergency_remount);
919  		schedule_work(work);
920  	}
921  }
922  
923  /*
924   * Unnamed block devices are dummy devices used by virtual
925   * filesystems which don't use real block-devices.  -- jrs
926   */
927  
928  static DEFINE_IDA(unnamed_dev_ida);
929  static DEFINE_SPINLOCK(unnamed_dev_lock);/* protects the above */
930  /* Many userspace utilities consider an FSID of 0 invalid.
931   * Always return at least 1 from get_anon_bdev.
932   */
933  static int unnamed_dev_start = 1;
934  
935  int get_anon_bdev(dev_t *p)
936  {
937  	int dev;
938  	int error;
939  
940   retry:
941  	if (ida_pre_get(&unnamed_dev_ida, GFP_ATOMIC) == 0)
942  		return -ENOMEM;
943  	spin_lock(&unnamed_dev_lock);
944  	error = ida_get_new_above(&unnamed_dev_ida, unnamed_dev_start, &dev);
945  	if (!error)
946  		unnamed_dev_start = dev + 1;
947  	spin_unlock(&unnamed_dev_lock);
948  	if (error == -EAGAIN)
949  		/* We raced and lost with another CPU. */
950  		goto retry;
951  	else if (error)
952  		return -EAGAIN;
953  
954  	if (dev >= (1 << MINORBITS)) {
955  		spin_lock(&unnamed_dev_lock);
956  		ida_remove(&unnamed_dev_ida, dev);
957  		if (unnamed_dev_start > dev)
958  			unnamed_dev_start = dev;
959  		spin_unlock(&unnamed_dev_lock);
960  		return -EMFILE;
961  	}
962  	*p = MKDEV(0, dev & MINORMASK);
963  	return 0;
964  }
965  EXPORT_SYMBOL(get_anon_bdev);
966  
967  void free_anon_bdev(dev_t dev)
968  {
969  	int slot = MINOR(dev);
970  	spin_lock(&unnamed_dev_lock);
971  	ida_remove(&unnamed_dev_ida, slot);
972  	if (slot < unnamed_dev_start)
973  		unnamed_dev_start = slot;
974  	spin_unlock(&unnamed_dev_lock);
975  }
976  EXPORT_SYMBOL(free_anon_bdev);
977  
978  int set_anon_super(struct super_block *s, void *data)
979  {
980  	return get_anon_bdev(&s->s_dev);
981  }
982  
983  EXPORT_SYMBOL(set_anon_super);
984  
985  void kill_anon_super(struct super_block *sb)
986  {
987  	dev_t dev = sb->s_dev;
988  	generic_shutdown_super(sb);
989  	free_anon_bdev(dev);
990  }
991  
992  EXPORT_SYMBOL(kill_anon_super);
993  
994  void kill_litter_super(struct super_block *sb)
995  {
996  	if (sb->s_root)
997  		d_genocide(sb->s_root);
998  	kill_anon_super(sb);
999  }
1000  
1001  EXPORT_SYMBOL(kill_litter_super);
1002  
1003  static int ns_test_super(struct super_block *sb, void *data)
1004  {
1005  	return sb->s_fs_info == data;
1006  }
1007  
1008  static int ns_set_super(struct super_block *sb, void *data)
1009  {
1010  	sb->s_fs_info = data;
1011  	return set_anon_super(sb, NULL);
1012  }
1013  
1014  struct dentry *mount_ns(struct file_system_type *fs_type,
1015  	int flags, void *data, void *ns, struct user_namespace *user_ns,
1016  	int (*fill_super)(struct super_block *, void *, int))
1017  {
1018  	struct super_block *sb;
1019  
1020  	/* Don't allow mounting unless the caller has CAP_SYS_ADMIN
1021  	 * over the namespace.
1022  	 */
1023  	if (!(flags & MS_KERNMOUNT) && !ns_capable(user_ns, CAP_SYS_ADMIN))
1024  		return ERR_PTR(-EPERM);
1025  
1026  	sb = sget_userns(fs_type, ns_test_super, ns_set_super, flags,
1027  			 user_ns, ns);
1028  	if (IS_ERR(sb))
1029  		return ERR_CAST(sb);
1030  
1031  	if (!sb->s_root) {
1032  		int err;
1033  		err = fill_super(sb, data, flags & MS_SILENT ? 1 : 0);
1034  		if (err) {
1035  			deactivate_locked_super(sb);
1036  			return ERR_PTR(err);
1037  		}
1038  
1039  		sb->s_flags |= MS_ACTIVE;
1040  	}
1041  
1042  	return dget(sb->s_root);
1043  }
1044  
1045  EXPORT_SYMBOL(mount_ns);
1046  
1047  #ifdef CONFIG_BLOCK
1048  static int set_bdev_super(struct super_block *s, void *data)
1049  {
1050  	s->s_bdev = data;
1051  	s->s_dev = s->s_bdev->bd_dev;
1052  
1053  	/*
1054  	 * We set the bdi here to the queue backing, file systems can
1055  	 * overwrite this in ->fill_super()
1056  	 */
1057  	s->s_bdi = bdev_get_queue(s->s_bdev)->backing_dev_info;
1058  	return 0;
1059  }
1060  
1061  static int test_bdev_super(struct super_block *s, void *data)
1062  {
1063  	return (void *)s->s_bdev == data;
1064  }
1065  
1066  struct dentry *mount_bdev(struct file_system_type *fs_type,
1067  	int flags, const char *dev_name, void *data,
1068  	int (*fill_super)(struct super_block *, void *, int))
1069  {
1070  	struct block_device *bdev;
1071  	struct super_block *s;
1072  	fmode_t mode = FMODE_READ | FMODE_EXCL;
1073  	int error = 0;
1074  
1075  	if (!(flags & MS_RDONLY))
1076  		mode |= FMODE_WRITE;
1077  
1078  	bdev = blkdev_get_by_path(dev_name, mode, fs_type);
1079  	if (IS_ERR(bdev))
1080  		return ERR_CAST(bdev);
1081  
1082  	/*
1083  	 * once the super is inserted into the list by sget, s_umount
1084  	 * will protect the lockfs code from trying to start a snapshot
1085  	 * while we are mounting
1086  	 */
1087  	mutex_lock(&bdev->bd_fsfreeze_mutex);
1088  	if (bdev->bd_fsfreeze_count > 0) {
1089  		mutex_unlock(&bdev->bd_fsfreeze_mutex);
1090  		error = -EBUSY;
1091  		goto error_bdev;
1092  	}
1093  	s = sget(fs_type, test_bdev_super, set_bdev_super, flags | MS_NOSEC,
1094  		 bdev);
1095  	mutex_unlock(&bdev->bd_fsfreeze_mutex);
1096  	if (IS_ERR(s))
1097  		goto error_s;
1098  
1099  	if (s->s_root) {
1100  		if ((flags ^ s->s_flags) & MS_RDONLY) {
1101  			deactivate_locked_super(s);
1102  			error = -EBUSY;
1103  			goto error_bdev;
1104  		}
1105  
1106  		/*
1107  		 * s_umount nests inside bd_mutex during
1108  		 * __invalidate_device().  blkdev_put() acquires
1109  		 * bd_mutex and can't be called under s_umount.  Drop
1110  		 * s_umount temporarily.  This is safe as we're
1111  		 * holding an active reference.
1112  		 */
1113  		up_write(&s->s_umount);
1114  		blkdev_put(bdev, mode);
1115  		down_write(&s->s_umount);
1116  	} else {
1117  		s->s_mode = mode;
1118  		snprintf(s->s_id, sizeof(s->s_id), "%pg", bdev);
1119  		sb_set_blocksize(s, block_size(bdev));
1120  		error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
1121  		if (error) {
1122  			deactivate_locked_super(s);
1123  			goto error;
1124  		}
1125  
1126  		s->s_flags |= MS_ACTIVE;
1127  		bdev->bd_super = s;
1128  	}
1129  
1130  	return dget(s->s_root);
1131  
1132  error_s:
1133  	error = PTR_ERR(s);
1134  error_bdev:
1135  	blkdev_put(bdev, mode);
1136  error:
1137  	return ERR_PTR(error);
1138  }
1139  EXPORT_SYMBOL(mount_bdev);
1140  
1141  void kill_block_super(struct super_block *sb)
1142  {
1143  	struct block_device *bdev = sb->s_bdev;
1144  	fmode_t mode = sb->s_mode;
1145  
1146  	bdev->bd_super = NULL;
1147  	generic_shutdown_super(sb);
1148  	sync_blockdev(bdev);
1149  	WARN_ON_ONCE(!(mode & FMODE_EXCL));
1150  	blkdev_put(bdev, mode | FMODE_EXCL);
1151  }
1152  
1153  EXPORT_SYMBOL(kill_block_super);
1154  #endif
1155  
1156  struct dentry *mount_nodev(struct file_system_type *fs_type,
1157  	int flags, void *data,
1158  	int (*fill_super)(struct super_block *, void *, int))
1159  {
1160  	int error;
1161  	struct super_block *s = sget(fs_type, NULL, set_anon_super, flags, NULL);
1162  
1163  	if (IS_ERR(s))
1164  		return ERR_CAST(s);
1165  
1166  	error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
1167  	if (error) {
1168  		deactivate_locked_super(s);
1169  		return ERR_PTR(error);
1170  	}
1171  	s->s_flags |= MS_ACTIVE;
1172  	return dget(s->s_root);
1173  }
1174  EXPORT_SYMBOL(mount_nodev);
1175  
1176  static int compare_single(struct super_block *s, void *p)
1177  {
1178  	return 1;
1179  }
1180  
1181  struct dentry *mount_single(struct file_system_type *fs_type,
1182  	int flags, void *data,
1183  	int (*fill_super)(struct super_block *, void *, int))
1184  {
1185  	struct super_block *s;
1186  	int error;
1187  
1188  	s = sget(fs_type, compare_single, set_anon_super, flags, NULL);
1189  	if (IS_ERR(s))
1190  		return ERR_CAST(s);
1191  	if (!s->s_root) {
1192  		error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
1193  		if (error) {
1194  			deactivate_locked_super(s);
1195  			return ERR_PTR(error);
1196  		}
1197  		s->s_flags |= MS_ACTIVE;
1198  	} else {
1199  		do_remount_sb(s, flags, data, 0);
1200  	}
1201  	return dget(s->s_root);
1202  }
1203  EXPORT_SYMBOL(mount_single);
1204  
1205  struct dentry *
1206  mount_fs(struct file_system_type *type, int flags, const char *name, void *data)
1207  {
1208  	struct dentry *root;
1209  	struct super_block *sb;
1210  	char *secdata = NULL;
1211  	int error = -ENOMEM;
1212  
1213  	if (data && !(type->fs_flags & FS_BINARY_MOUNTDATA)) {
1214  		secdata = alloc_secdata();
1215  		if (!secdata)
1216  			goto out;
1217  
1218  		error = security_sb_copy_data(data, secdata);
1219  		if (error)
1220  			goto out_free_secdata;
1221  	}
1222  
1223  	root = type->mount(type, flags, name, data);
1224  	if (IS_ERR(root)) {
1225  		error = PTR_ERR(root);
1226  		goto out_free_secdata;
1227  	}
1228  	sb = root->d_sb;
1229  	BUG_ON(!sb);
1230  	WARN_ON(!sb->s_bdi);
1231  	sb->s_flags |= MS_BORN;
1232  
1233  	error = security_sb_kern_mount(sb, flags, secdata);
1234  	if (error)
1235  		goto out_sb;
1236  
1237  	/*
1238  	 * filesystems should never set s_maxbytes larger than MAX_LFS_FILESIZE
1239  	 * but s_maxbytes was an unsigned long long for many releases. Throw
1240  	 * this warning for a little while to try and catch filesystems that
1241  	 * violate this rule.
1242  	 */
1243  	WARN((sb->s_maxbytes < 0), "%s set sb->s_maxbytes to "
1244  		"negative value (%lld)\n", type->name, sb->s_maxbytes);
1245  
1246  	up_write(&sb->s_umount);
1247  	free_secdata(secdata);
1248  	return root;
1249  out_sb:
1250  	dput(root);
1251  	deactivate_locked_super(sb);
1252  out_free_secdata:
1253  	free_secdata(secdata);
1254  out:
1255  	return ERR_PTR(error);
1256  }
1257  
1258  /*
1259   * This is an internal function, please use sb_end_{write,pagefault,intwrite}
1260   * instead.
1261   */
1262  void __sb_end_write(struct super_block *sb, int level)
1263  {
1264  	percpu_up_read(sb->s_writers.rw_sem + level-1);
1265  }
1266  EXPORT_SYMBOL(__sb_end_write);
1267  
1268  /*
1269   * This is an internal function, please use sb_start_{write,pagefault,intwrite}
1270   * instead.
1271   */
1272  int __sb_start_write(struct super_block *sb, int level, bool wait)
1273  {
1274  	bool force_trylock = false;
1275  	int ret = 1;
1276  
1277  #ifdef CONFIG_LOCKDEP
1278  	/*
1279  	 * We want lockdep to tell us about possible deadlocks with freezing
1280  	 * but it's it bit tricky to properly instrument it. Getting a freeze
1281  	 * protection works as getting a read lock but there are subtle
1282  	 * problems. XFS for example gets freeze protection on internal level
1283  	 * twice in some cases, which is OK only because we already hold a
1284  	 * freeze protection also on higher level. Due to these cases we have
1285  	 * to use wait == F (trylock mode) which must not fail.
1286  	 */
1287  	if (wait) {
1288  		int i;
1289  
1290  		for (i = 0; i < level - 1; i++)
1291  			if (percpu_rwsem_is_held(sb->s_writers.rw_sem + i)) {
1292  				force_trylock = true;
1293  				break;
1294  			}
1295  	}
1296  #endif
1297  	if (wait && !force_trylock)
1298  		percpu_down_read(sb->s_writers.rw_sem + level-1);
1299  	else
1300  		ret = percpu_down_read_trylock(sb->s_writers.rw_sem + level-1);
1301  
1302  	WARN_ON(force_trylock && !ret);
1303  	return ret;
1304  }
1305  EXPORT_SYMBOL(__sb_start_write);
1306  
1307  /**
1308   * sb_wait_write - wait until all writers to given file system finish
1309   * @sb: the super for which we wait
1310   * @level: type of writers we wait for (normal vs page fault)
1311   *
1312   * This function waits until there are no writers of given type to given file
1313   * system.
1314   */
1315  static void sb_wait_write(struct super_block *sb, int level)
1316  {
1317  	percpu_down_write(sb->s_writers.rw_sem + level-1);
1318  }
1319  
1320  /*
1321   * We are going to return to userspace and forget about these locks, the
1322   * ownership goes to the caller of thaw_super() which does unlock().
1323   */
1324  static void lockdep_sb_freeze_release(struct super_block *sb)
1325  {
1326  	int level;
1327  
1328  	for (level = SB_FREEZE_LEVELS - 1; level >= 0; level--)
1329  		percpu_rwsem_release(sb->s_writers.rw_sem + level, 0, _THIS_IP_);
1330  }
1331  
1332  /*
1333   * Tell lockdep we are holding these locks before we call ->unfreeze_fs(sb).
1334   */
1335  static void lockdep_sb_freeze_acquire(struct super_block *sb)
1336  {
1337  	int level;
1338  
1339  	for (level = 0; level < SB_FREEZE_LEVELS; ++level)
1340  		percpu_rwsem_acquire(sb->s_writers.rw_sem + level, 0, _THIS_IP_);
1341  }
1342  
1343  static void sb_freeze_unlock(struct super_block *sb)
1344  {
1345  	int level;
1346  
1347  	for (level = SB_FREEZE_LEVELS - 1; level >= 0; level--)
1348  		percpu_up_write(sb->s_writers.rw_sem + level);
1349  }
1350  
1351  /**
1352   * freeze_super - lock the filesystem and force it into a consistent state
1353   * @sb: the super to lock
1354   *
1355   * Syncs the super to make sure the filesystem is consistent and calls the fs's
1356   * freeze_fs.  Subsequent calls to this without first thawing the fs will return
1357   * -EBUSY.
1358   *
1359   * During this function, sb->s_writers.frozen goes through these values:
1360   *
1361   * SB_UNFROZEN: File system is normal, all writes progress as usual.
1362   *
1363   * SB_FREEZE_WRITE: The file system is in the process of being frozen.  New
1364   * writes should be blocked, though page faults are still allowed. We wait for
1365   * all writes to complete and then proceed to the next stage.
1366   *
1367   * SB_FREEZE_PAGEFAULT: Freezing continues. Now also page faults are blocked
1368   * but internal fs threads can still modify the filesystem (although they
1369   * should not dirty new pages or inodes), writeback can run etc. After waiting
1370   * for all running page faults we sync the filesystem which will clean all
1371   * dirty pages and inodes (no new dirty pages or inodes can be created when
1372   * sync is running).
1373   *
1374   * SB_FREEZE_FS: The file system is frozen. Now all internal sources of fs
1375   * modification are blocked (e.g. XFS preallocation truncation on inode
1376   * reclaim). This is usually implemented by blocking new transactions for
1377   * filesystems that have them and need this additional guard. After all
1378   * internal writers are finished we call ->freeze_fs() to finish filesystem
1379   * freezing. Then we transition to SB_FREEZE_COMPLETE state. This state is
1380   * mostly auxiliary for filesystems to verify they do not modify frozen fs.
1381   *
1382   * sb->s_writers.frozen is protected by sb->s_umount.
1383   */
1384  int freeze_super(struct super_block *sb)
1385  {
1386  	int ret;
1387  
1388  	atomic_inc(&sb->s_active);
1389  	down_write(&sb->s_umount);
1390  	if (sb->s_writers.frozen != SB_UNFROZEN) {
1391  		deactivate_locked_super(sb);
1392  		return -EBUSY;
1393  	}
1394  
1395  	if (!(sb->s_flags & MS_BORN)) {
1396  		up_write(&sb->s_umount);
1397  		return 0;	/* sic - it's "nothing to do" */
1398  	}
1399  
1400  	if (sb->s_flags & MS_RDONLY) {
1401  		/* Nothing to do really... */
1402  		sb->s_writers.frozen = SB_FREEZE_COMPLETE;
1403  		up_write(&sb->s_umount);
1404  		return 0;
1405  	}
1406  
1407  	sb->s_writers.frozen = SB_FREEZE_WRITE;
1408  	/* Release s_umount to preserve sb_start_write -> s_umount ordering */
1409  	up_write(&sb->s_umount);
1410  	sb_wait_write(sb, SB_FREEZE_WRITE);
1411  	down_write(&sb->s_umount);
1412  
1413  	/* Now we go and block page faults... */
1414  	sb->s_writers.frozen = SB_FREEZE_PAGEFAULT;
1415  	sb_wait_write(sb, SB_FREEZE_PAGEFAULT);
1416  
1417  	/* All writers are done so after syncing there won't be dirty data */
1418  	sync_filesystem(sb);
1419  
1420  	/* Now wait for internal filesystem counter */
1421  	sb->s_writers.frozen = SB_FREEZE_FS;
1422  	sb_wait_write(sb, SB_FREEZE_FS);
1423  
1424  	if (sb->s_op->freeze_fs) {
1425  		ret = sb->s_op->freeze_fs(sb);
1426  		if (ret) {
1427  			printk(KERN_ERR
1428  				"VFS:Filesystem freeze failed\n");
1429  			sb->s_writers.frozen = SB_UNFROZEN;
1430  			sb_freeze_unlock(sb);
1431  			wake_up(&sb->s_writers.wait_unfrozen);
1432  			deactivate_locked_super(sb);
1433  			return ret;
1434  		}
1435  	}
1436  	/*
1437  	 * For debugging purposes so that fs can warn if it sees write activity
1438  	 * when frozen is set to SB_FREEZE_COMPLETE, and for thaw_super().
1439  	 */
1440  	sb->s_writers.frozen = SB_FREEZE_COMPLETE;
1441  	lockdep_sb_freeze_release(sb);
1442  	up_write(&sb->s_umount);
1443  	return 0;
1444  }
1445  EXPORT_SYMBOL(freeze_super);
1446  
1447  /**
1448   * thaw_super -- unlock filesystem
1449   * @sb: the super to thaw
1450   *
1451   * Unlocks the filesystem and marks it writeable again after freeze_super().
1452   */
1453  int thaw_super(struct super_block *sb)
1454  {
1455  	int error;
1456  
1457  	down_write(&sb->s_umount);
1458  	if (sb->s_writers.frozen != SB_FREEZE_COMPLETE) {
1459  		up_write(&sb->s_umount);
1460  		return -EINVAL;
1461  	}
1462  
1463  	if (sb->s_flags & MS_RDONLY) {
1464  		sb->s_writers.frozen = SB_UNFROZEN;
1465  		goto out;
1466  	}
1467  
1468  	lockdep_sb_freeze_acquire(sb);
1469  
1470  	if (sb->s_op->unfreeze_fs) {
1471  		error = sb->s_op->unfreeze_fs(sb);
1472  		if (error) {
1473  			printk(KERN_ERR
1474  				"VFS:Filesystem thaw failed\n");
1475  			lockdep_sb_freeze_release(sb);
1476  			up_write(&sb->s_umount);
1477  			return error;
1478  		}
1479  	}
1480  
1481  	sb->s_writers.frozen = SB_UNFROZEN;
1482  	sb_freeze_unlock(sb);
1483  out:
1484  	wake_up(&sb->s_writers.wait_unfrozen);
1485  	deactivate_locked_super(sb);
1486  	return 0;
1487  }
1488  EXPORT_SYMBOL(thaw_super);
1489